Democracy and Demography:Societal Effects of Fertility Limits on Local Leaders

S AnukritiBoston College

Abhishek ChakravartyUniversity of Essex

1 / 43

Introduction

• Representative democracy is good for welfare:

• Promotes stable economic growth (Rodrik (2000), Mobarak (2005))

• Narrows income inequalities (Acemoglu and Robinson (2000), Gradstein (2007))

• Prevents elite capture (Foster and Rosenzweig (2004), Brown and Mobarak (2009))

• Wider candidate pool → better quality leaders in democratic systems(Besley and Coate (1997), Besley and Reynal-Querol (2011), Osborne and Slivinski (1996))

• Typically, minimal legal restrictions on who can become an electeddemocratic leader

2 / 43

• In practice, democratic leaders may be of “poor” quality:• Substantial entry barriers to the candidate pool

– political networks, campaign costs, and other socioeconomic inequities• Voters may have imperfect information on candidates’ characteristics• Voters may prefer to elect leaders who can provide patronage at the expense of

other constituents

• Some countries have sought to improve candidate quality by imposing“desirable” characteristics on candidates, such as minimum education levelsand no criminal convictions

• Limited evidence on the effects of these “quality controls” on policyoutcomes and citizens’ behavior

3 / 43

• In practice, democratic leaders may be of “poor” quality:• Substantial entry barriers to the candidate pool

– political networks, campaign costs, and other socioeconomic inequities• Voters may have imperfect information on candidates’ characteristics• Voters may prefer to elect leaders who can provide patronage at the expense of

other constituents

• Some countries have sought to improve candidate quality by imposing“desirable” characteristics on candidates, such as minimum education levelsand no criminal convictions

• Limited evidence on the effects of these “quality controls” on policyoutcomes and citizens’ behavior

3 / 43

This paper

• We focus on a unique policy experiment in India

• Since 1992, several Indian states bar individuals (male or female) with morethan 2 children from contesting village council (Panchayat) elections

• First instance of a democratic country instituting a fertility ceiling forelectoral candidates

• We examine the impacts of the two-child limits on fertility outcomes amongthe constituents/ general population

• Can restricting elected leadership positions to candidates with “desirable”attributes lead citizens to adopt those attributes?Yes, but...

4 / 43

This paper

• We focus on a unique policy experiment in India

• Since 1992, several Indian states bar individuals (male or female) with morethan 2 children from contesting village council (Panchayat) elections

• First instance of a democratic country instituting a fertility ceiling forelectoral candidates

• We examine the impacts of the two-child limits on fertility outcomes amongthe constituents/ general population

• Can restricting elected leadership positions to candidates with “desirable”attributes lead citizens to adopt those attributes?

Yes, but...

4 / 43

This paper

• We focus on a unique policy experiment in India

• Since 1992, several Indian states bar individuals (male or female) with morethan 2 children from contesting village council (Panchayat) elections

• First instance of a democratic country instituting a fertility ceiling forelectoral candidates

• We examine the impacts of the two-child limits on fertility outcomes amongthe constituents/ general population

• Can restricting elected leadership positions to candidates with “desirable”attributes lead citizens to adopt those attributes?Yes, but...

4 / 43

Policy Details

• Starting in 1992, 11 states have enacted fertility limits for Panchayats and/or municipal elections

• 4 states revoked them in 2005, but they remain in effect in 7 states

• One year grace-period: births during this year were exempt

• If ≥ 2 children when the law was announced:• Additional births after the grace-period =⇒ disqualification

• If < 2 children when the law was announced:• Third birth after the grace-period =⇒ disqualification

Disqualifications

5 / 43

Background: Panchayats

• India is a stable democracy

• Panchayats are the lowest unit of governance in India

• Granted constitutional status in 1992

• 3 tiers: village councils, block councils, and district councils

• Regular elections every 5 years

• No term limits on Panchayat members

• Minimum age to contest elections is 21 years

• Average population per village Panchayat ≈ 3,100

• Voter turnout in Panchayat elections generally > 70%

• Gender and caste quotas

6 / 43

Background: Fertility limits

• India is world’s second most populous country and houses one-third of itspoorest

• This manipulation of the candidate pool aims to curb population growth,and is not intended to directly improve leaders’ performance

• Seek to improve economic outcomes by precipitating fertility decline

• Stated mechanism: role-model channel and by conveying policymakers’seriousness about curbing population growth

• The limits, however, also incentivize individuals who intend to contestelections to plan smaller families

• May lead to fear or anticipation of stricter fertility limits in otherdimensions, such as for government jobs

7 / 43

Background: Fertility limits

• India is world’s second most populous country and houses one-third of itspoorest

• This manipulation of the candidate pool aims to curb population growth,and is not intended to directly improve leaders’ performance

• Seek to improve economic outcomes by precipitating fertility decline

• Stated mechanism: role-model channel and by conveying policymakers’seriousness about curbing population growth

• The limits, however, also incentivize individuals who intend to contestelections to plan smaller families

• May lead to fear or anticipation of stricter fertility limits in otherdimensions, such as for government jobs

7 / 43

Policy Relevance

• The limits impact a large share of potential candidates of childbearing age– In Rajasthan, 47% of Panchayat members in 2012 under 36 years of age and 41% in the36-50 year age-group

• Official salaries of Panchayat members are not substantial– typical monthly salary of a village council head is about USD 50 - USD 60

• However, Panchayats have considerable power at the local level andmembers have discretion over a large share of local funds

• Receive substantial funds from the national and state governments and areauthorized to implement development schemes, e.g., NREGA

• Can collect taxes, license fees, and fines, and receive seignorage from the auction oflocal mineral and forestry resources

• Responsible for provision of public goods, such as roads and wells

• High potential private returns from political rents and corrupt practices

8 / 43

Policy Relevance

• The limits impact a large share of potential candidates of childbearing age– In Rajasthan, 47% of Panchayat members in 2012 under 36 years of age and 41% in the36-50 year age-group

• Official salaries of Panchayat members are not substantial– typical monthly salary of a village council head is about USD 50 - USD 60

• However, Panchayats have considerable power at the local level andmembers have discretion over a large share of local funds

• Receive substantial funds from the national and state governments and areauthorized to implement development schemes, e.g., NREGA

• Can collect taxes, license fees, and fines, and receive seignorage from the auction oflocal mineral and forestry resources

• Responsible for provision of public goods, such as roads and wells

• High potential private returns from political rents and corrupt practices

8 / 43

Policy Relevance

• The limits impact a large share of potential candidates of childbearing age– In Rajasthan, 47% of Panchayat members in 2012 under 36 years of age and 41% in the36-50 year age-group

• Official salaries of Panchayat members are not substantial– typical monthly salary of a village council head is about USD 50 - USD 60

• However, Panchayats have considerable power at the local level andmembers have discretion over a large share of local funds

• Receive substantial funds from the national and state governments and areauthorized to implement development schemes, e.g., NREGA

• Can collect taxes, license fees, and fines, and receive seignorage from the auction oflocal mineral and forestry resources

• Responsible for provision of public goods, such as roads and wells

• High potential private returns from political rents and corrupt practices

8 / 43

Data

• Pool 3 repeated cross-sections of National Family Health Survey-1,2,3

• Years of survey: 1992-93, 1998-99, 2005-06

• Each round is representative at the state-level

• Complete birth history for each woman

• e.g., month and year of child’s birth, birth order, mother’s age at birth

• Construct a large, retrospective, unbalanced woman-year panel

• Entry in the year of marriage and exit in the year of survey

• Sample period: 1973-2006– We cannot credibly examine the effect of revocations that took place in 2005

• 99,804 women and 256,267 births from 18 states

9 / 43

Timeline

State Announced Grace Period End

Rajasthan 1992 Apr 23, 1994 - Nov 27, 1995Haryana 1994 Apr 21, 1994 - Apr 24, 1995 Jul 21, 2006Andhra Pradesh 1994 May 30, 1994 - May 30, 1995Orissa 1993/1994* Apr 1994 - Apr 21, 1995Himachal Pradesh 2000 Apr 18, 2000 - Apr 18, 2001 Apr 5, 2005Madhya Pradesh 2000** Mar 29, 2000 - Jan 26, 2001 Nov 20, 2005Chhattisgarh 2000 2000 - Jan 2001 2005Maharashtra 2003*** Sep 21, 2002 - Sep 20, 2003Uttarakhand (municipal) 2002Gujarat 2005 Aug 2005 - Aug 11, 2006Bihar (municipal) Jan 2007 Feb 1, 2007 - Feb 1, 2008

*For district councils in 1993 and for village and block councils in 1994.

**Notified on May 31, 2000. This created problems since people whose third child was born in Jan 2001 contested theirdisqualification for birth within 8 months of the new law.***In retrospective effect from Sep 21, 2002.

11 / 43

Emprical Strategy

• Goal: to estimate the causal effect of the two-child limits on candidates invillage council elections in a state on fertility-related outcomes amongresidents in the same state

• Utilize the quasi-experimental geographical and temporal variation inannouncement of these laws across Indian states

• We can estimate the impact for only 7 (8) “treated” states– Rajasthan, Haryana, AP, Orissa, HP, MP (including Chhattisgarh), and Maharashtra– The limits came into effect in Bihar and Gujarat after 2006

• 9 control states

• Treatment year is based on the year of announcement of the law

12 / 43

Treatment and Control States

13 / 43

Event-Study Hazard Analysis

• Evolution of the hazard of birth before and after the laws were announced

• For a woman i of age a in state s and year t:

Yiast =

5∑k=−6

αkTs∗Posts,t+k+

5∑k=−6

βkPosts,t+k+X′i δ+γs+θt+ψa+νs∗t+ϵiast (1)

• Ts indicates treatment states

• For treatment states, Posts,t+k indicates years during which the law is inplace in state s– The year before the year of announcement is the omitted year

• For never-treated states, we use fictitious announcement years– Assign the same announcement year to a control state as its neighboring treatment state– If multiple bordering T states, we randomly choose the T year of one of the neighbors– Results robust to alternate assignments of placebo announcement years

15 / 43

• Outcomes are indicators for 1st/ 2nd/ 3rd/ 4th/ 5th birth

• For hazard of birth b, sample restricted to years after birth (b − 1), until b,and to women whose previous (b − 1) children were born beforeannouncement of the law in their respective states

• Xi: woman’s and her husband’s years of schooling, indicators for religion,caste, standard of living, urban residence, year of interview

• If there is no noticeable pre-trend in the differential hazard of birth acrosstreatment and control states, we can interpret the αk coefficients during thepost-announcement years as the causal effect of the limits

• k ≤ 5 to equalize the number of post-treatment years across states

16 / 43

Net effect on birth hazards

We pool the event study coefficients in (1) and estimate:

Yiast = ω+αTs ∗ Postst + βPostst +X′

i δ+ γs + θt +ψa + νs ∗ t+ µsa + ϵiast (2a)

Yiast = ω + αTreatst + X′

i δ + γs + θt + ψa + νs ∗ t + µsa + ϵiast (2b)

• (2a) corresponds to (1)

• In (2b), we define treatment as zero for all control states:– Treatst = 1 for women in treated states if t ≥ the year of announcement, and zero o.w.

• Use all available pre- and post-announcement years for each state

• Also control for years since last birth (marriage) flexibly

• (1)-(2b) capture the effects on marginal fertility of affected households

17 / 43

Effect on Total Number of Living Children

• Re-estimate (2a) and (2b) using indicators for whether a woman i of age ain state s and year t reports having 1/ 2/ 3/ 4/ 5 living children in year t asthe outcome variables

• Unlike the hazard analysis, no restrictions in terms of the prior number ofchildren

• Use all available years for each woman

• If the two-child limits are effective, we expect the likelihood of having 2children to increase in the treatment states after the laws have beenannounced

• Capture the marginal effects on couples who had begun childbearing beforethe laws were announced + the behavioral response of new parents whobegan childbearing post-announcement

18 / 43

Sex Ratio of births

• The two-child laws may also affect the sex ratio of births• We examine the effect on sex ratio of second and higher order births• Despite the availability of prenatal sex-determination technology, sex of the

first birth is random in India(Bhalotra and Cochrane (2010), Dasgupta and Bhat (1997), Visaria (2005))

• Parents more likely to practice sex-selection at higher parities if they do nothave a son (Portner (2010), Rosenblum (2013), Anukriti et al. (2016))

• Restrict sample to women whose 1st child was born before theannouncement–Sex ratio at 1st birth is “normal” in control and treatment states (both pre and post)

19 / 43

Maleiast = α+ βTreatst + X′i δ + γs + θt + ψa + νs ∗ t + µsa + ϕGirli + ϵiast (3a)

Maleiast = α+β1Ts∗Postst+β2Postst+X′i δ+γs+θt+ψa+νs∗t+µsa+ϕGirli+ϵiast (3b)

• Outcome variable: child is male• Girli: mother i’s first child is a girl

20 / 43

Identifying assumption

• The state-year variation in the timing of law announcement is uncorrelatedwith other time-varying determinants of the outcomes of interest.

• Test for correlations between law announcements and HH characteristics:

SESist = α+ βTreatst + γs + θt + νs ∗ t + ϵist

• To the best of our knowledge, during the time-period we examine, therewere no other state-specific programs in the treatment states that promotedsmaller families and whose timing coincided with the fertility limits

22 / 43

Coefficient of Treatst Std. ErrorDependent Variable (1) (2)

SC -0.004 [0.008]ST 0.009 [0.008]OBC -0.008 [0.010]Upper caste 0.003 [0.011]Hindu 0.012 [0.009]Muslim 0.003 [0.006]Sikh 0.001 [0.002]Christian 0.001 [0.007]Low SLI 0.009 [0.008]Med SLI -0.001 [0.006]High SLI -0.007 [0.005]Wife’s years of schooling:Zero -0.005 [0.007]5-10 years 0.009 [0.010]10-12 years 0.002 [0.002]12-15 years 0.001 [0.004]≥ 15 years -0.002 [0.002]Husband’s years of schooling:Zero 0.003 [0.008]5-10 years -0.002 [0.008]10-12 years -0.001 [0.003]12-15 years 0.002 [0.005]≥ 15 years -0.000 [0.003]

N 1,143,05723 / 43

Differences in birth hazards for T and C states (αk)

24 / 43

Net effect on the hazard of third birth

Declines by 10-11%

3rd birth = 1 (1) (2) (3)

Panel A:Treatst -0.0143 -0.0213 -0.0206

[0.0096] [0.0088]** [0.0078]**(0.0095) (0.0098)** (0.0093)**

Baseline mean 0.2131

Panel B:Ts ∗ Postst -0.0196 -0.0265 -0.0263

[0.0114] [0.0117]** [0.0103]**(0.0117) (0.0131)* (0.0120)**

Postst 0.0099 0.0077 0.0083[0.0101] [0.0090] [0.0082]

Baseline mean 0.2431

N 182,082

State FE x x xYear FE x x xYears since 2nd birth FE x x xXit x x xLinear state trends x xState x Age FE x

NOTES: This table reports the coefficients from specifications (2a) and (2b). Standard errors in brackets are clustered by stateand in parentheses are wild-cluster bootstrapped by state. *** 1%, ** 5%, * 10%. 26 / 43

Heterogeneity in the net effect on the hazard of third birth

3rd birth = 1 SC ST OBC Upper Low SLI High SLI Wife has Wife has Husband has Husband hasschooling no schooling schooling no schooling

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Panel A:Treatst -0.0410 0.0029 -0.0288 -0.0098 -0.0214 -0.0024 -0.0160 -0.0182 -0.0159 -0.0294

[0.0126] *** [0.0215] [0.0104]** [0.0124] [0.0100]** [0.0123] [0.0103] [0.0097]* [0.0077]* [0.0119]**(0.0168)*** (0.0207) (0.0142)* (0.0119) (0.0098)* (0.0119) (0.0095) (0.0103) (0.0083)* (0.0146)**

Baseline mean 0.2475 0.2521 0.1990 0.2015 0.2478 0.1143 0.1468 0.2629 0.1917 0.2604

Panel B:Ts ∗ Postst -0.0467 -0.0094 -0.0252 -0.0182 -0.0296 -0.0062 -0.0227 -0.0246 -0.0221 -0.0336

[0.0157]*** [0.0314] [0.0117]** [0.0162] [0.0123]** [0.0131] [0.0114]* [0.0119]* [0.0099]** [0.0171]*(0.0180)*** (0.0300) (0.0132)* (0.0162) (0.0133)** (0.0127) (0.0119)* (0.0126)* (0.0107)** (0.0178)*

Postst 0.0083 0.0171 -0.0046 0.0125 0.0119 0.0054 0.0097 0.0090 0.0090 0.0058[0.0110] [0.0286] [0.0114] [0.0122] [0.0113] [0.0100] [0.0073] [0.0114] [0.0076] [0.0147]

Baseline mean 0.2831 0.2690 0.2470 0.2257 0.2669 0.1543 0.1830 0.2777 0.2257 0.2769

N 28,074 17,868 38,288 97,852 110,159 17,200 72,165 109,917 122,979 59,103

NOTES: This table reports the coefficients from specifications (2a) and (2b). Standard errors in brackets are clustered by stateand in parentheses are wild-cluster bootstrapped by state. The baseline mean is calculated for observations where Treatst = 0

in panel A and for observations where Postst = 0 in panel B. *** 1%, ** 5%, * 10%.

28 / 43

Effects on the number of living children

Kids = 1 Kids = 2 Kids = 3 Kids = 4 Kids = 5(1) (2) (3) (4) (5)

Panel A: Only treatment statesTreatst 0.0066 0.0075 -0.0042 -0.0047 -0.0028

[0.0039] [0.0035]* [0.0021]* [0.0025] [0.0013]*(0.0046) (0.0042)* (0.0023)* (0.0030) (0.0017)*

N 459,293Baseline mean 0.2394 0.2199 0.1693 0.0836 0.0322

Panel B:Treatst 0.0008 0.0090 -0.0018 -0.0052 -0.0024

[0.0055] [0.0068] [0.0055] [0.0026]* [0.0020](0.0054) (0.0065) (0.0053) (0.0030)* (0.0053)

N 1,143,057Baseline mean 0.2351 0.2351 0.1711 0.0878 0.0379

Panel C:Ts ∗ Postst 0.0001 0.0139 -0.0009 -0.0080 -0.0014

[0.0063] [0.0100] [0.0073] [0.0037]** [0.0026](0.0062) (0.0104) (0.0072) (0.0042)* (0.0026)

Postst 0.0009 -0.0067 -0.0013 0.0040 -0.0014[0.0043] [0.0062] [0.0037] [0.0025] [0.0026]

N 1,143,057Baseline mean 0.2425 0.2220 0.1650 0.0855 0.0364

NOTES: This table reports the coefficients from specifications (2a) and (2b). Standard errors in brackets are clustered by stateand in parentheses are wild-cluster bootstrapped by state. *** 1%, ** 5%, * 10%. Pr(2 kids) ↑ by 3.41%.

29 / 43

Heterogeneous effects on the likelihood of > 2 living children

Pr(>2 kids) declines by 4.42% in treatment states

All SC ST OBC Upper Low SLI High SLI Wife has Wife has Husband has Husband hasschooling no schooling schooling no schooling

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

A. Treatment states onlyTreatst -0.0133 0.0045 -0.0402 -0.0068 -0.0168 -0.0190 -0.0109 -0.0110 -0.0155 -0.0136 -0.0131

[0.0043]** [0.0126] [0.0185]* [0.0062] [0.0118] [0.0088]* [0.0039]** [0.0037]** [0.0078]* [0.0033]*** [0.0097](0.0072)** (0.0119) (0.0276)** (0.0062) (0.0111) (0.0135) (0.0062)* (0.0057)** (0.0115) (0.0063)*** (0.0125)

N 459,293 78,174 77,278 105,475 198,366 291,535 33,708 149,776 309,517 292,311 166,982Baseline mean 0.3007 0.3124 0.3013 0.2860 0.3023 0.3064 0.2561 0.2674 0.3152 0.2938 0.3124

B. All statesTreatst -0.0087 0.00002 -0.0206 -0.0097 -0.0141 -0.0100 -0.0111 -0.0121 -0.0081 -0.0110 -0.0049

[0.0065] [0.0108] [0.0147] [0.0068] [0.0129] [0.0064] [0.0062]* [0.0069] [0.0063] [0.0070] [0.0067](0.0069) (0.0104) (0.0194) (0.0072) (0.0133) (0.0077) (0.0074) (0.0077) (0.0073) (0.0075) (0.0073)

N 1,143,057 202,619 123,071 267,024 550,343 722,793 90,528 416,265 726,792 747,865 395,192Baseline mean 0.3189 0.3399 0.3180 0.3129 0.3144 0.3317 0.2498 0.2592 0.3527 0.3027 0.3495

NOTES: This table reports the coefficients from specifications (2a) and (2b). Standard errors in brackets are clustered by stateand in parentheses are wild-cluster bootstrapped by state. *** 1%, ** 5%, * 10%.

30 / 43

Sex ratio of second and higher parity births

Male = 1 All SC ST OBC Upper(1) (2) (3) (4) (5)

Panel A: Only treatment statesTreatst 0.0086 -0.0265 0.0071 0.0528 -0.0048

[0.0103] [0.0419] [0.0082] [0.0111]*** [0.0210](0.0111) (0.0368) (0.0066) (0.0252)** (0.0196)

N 61,490 11,054 11,627 12,677 26,132Baseline mean 0.5211 0.5235 0.5142 0.5117 0.5267

Panel B:Treatst 0.0109 -0.0249 0.0071 0.0557 0.0061

[0.0060]* [0.0232] [0.0157] [0.0119]*** [0.0148](0.0078) (0.0218) (0.0154) (0.0221)** (0.0143)

N 165,016 31,169 18,757 35,858 79,232Baseline mean 0.5186 0.5215 0.5177 0.5185 0.5178

Panel C:Ts ∗ Postst 0.0088 -0.0231 -0.0137 0.0570 0.0022

[0.0078] [0.0208] [0.0241] [0.0145]*** [0.0168](0.0077) (0.0201) (0.0241) (0.0202)*** (0.0164)

Postst 0.0032 -0.0026 0.0288 -0.0017 0.0060[0.0082] [0.0144] [0.0272] [0.0100] [0.0123]

N 165,016 31,169 18,757 35,858 79,232Baseline mean 0.5181 0.5214 0.5174 0.5184 0.5170

NOTES: This table reports the coefficients from specifications (3a) and (3b). Standard errors in brackets are clustered by stateand in parentheses are wild-cluster bootstrapped by state. *** 1%, ** 5%, * 10%. 10.32% ↑ for OBCs.

Sex ratio of first births32 / 43

• Upper castes are “less treatable” because of low fertility at baseline• Consistent with lack of significant fertility effects for upper castes• OBCs constitute significant fractions of the populations in our treatment

states, such as Haryana (28.1%), Rajasthan (47.5%), Madhya Pradesh(41.2%), and Maharashtra (27.1%) that have highly adverse sex ratios

33 / 43

Robustness 1: Synthetic Control

•

• Abadie and Gardeazabal (2003) and Abadie et al. (2010)

34 / 43

Robustness 2: Alternate placebo years for control states

Treatment year assigned to control states:1993 1994 1995 1996 1997 1998 1999

3rd birth = 1 (1) (2) (3) (4) (5) (6) (7)

Panel A:Treatst -0.0205 -0.0200 -0.0212 -0.0232 -0.0241 -0.0244 -0.0223

[0.0084]** [0.0085]** [0.0083]** [0.0082]** [0.0082]*** [0.0083]*** [0.0086]**(0.0102)** (0.0098)* (0.0097)** (0.0098)** (0.0101)** (0.0102)** (0.0103)**

Panel B:Ts ∗ Postst -0.0222 -0.0210 -0.0213 -0.0238 -0.0230 -0.0221 -0.0341

[0.0099]** [0.0092]** [0.0122]* [0.0120]* [0.0135] [0.0166] [0.0144]**(0.0118)** (0.0102)** (0.0129) (0.0132)* (0.0152) (0.0181) (0.0170)**

Postst 0.0028 0.0015 0.0001 0.0007 -0.0013 -0.0030 0.0177[0.0115] [0.0105] [0.0127] [0.0117] [0.0117] [0.0153] [0.0130]

N 164,843 171,975 178,584 184,751 190,071 195,029 199,100

NOTES: Standard errors in brackets are clustered by state and in parentheses are wild-cluster bootstrapped by state. *** 1%,** 5%, * 10%.

35 / 43

Robustness 3: Use all neighbors as control states

Match each T state with all its neighboring C states and create a new dataset in which C states

that border multiple T states appear multiple times

3rd birth = 1 (1) (2) (3)

Panel A:Treatst -0.0149 -0.0199 -0.0188

[0.0097] [0.0088]** [0.0079]**(0.0093) (0.0103)* (0.0090)**

Panel B:Ts ∗ Postst -0.0170 -0.0216 -0.0217

[0.0100] [0.0092]** [0.0083]**(0.0100) (0.0107)* (0.0097)**

Postst 0.0061 0.0031 0.0051[0.0042] [0.0035] [0.0035]

N 292,514

State FE x x xYear FE x x xYears since 2nd birth FE x x xXit x x xLinear state trends x xState x Age FE x

NOTES: Standard errors in brackets are clustered by state and in parentheses are wild-cluster bootstrapped by state. *** 1%,** 5%, * 10%. We weight each observation by the square root of the inverse of the number of times an observation appears inthe sample. 36 / 43

Magnitudes

• Average baseline terminal fertility in the treatment states is 2.8– 30% have > 2 children

• We find that 615,390 (1.33% of) rural couples ↓ fertility due to the limits

• Hazard of 3rd birth estimates =⇒ 2.65% of rural couples who had twochildren in T states ↓ marginal fertility due to the limits

• Comparison with other program impacts:

• Matlab FP interventions ↓ fertility by 17-23% (Canning and Schultz (2012))

• China’s OCP ↓ fertility by 2% (Almond et al (2013))

• Devi Rupak in India ↓ fertility by 1% (Anukriti (2014))

• Our estimated effects fall in between

37 / 43

Mechanisms

• Our estimated impacts are consistent with the high participation of votersand candidates in local politics, making both the aspirations and role-modelchannels plausible

• 2014 World Values Survey for India: 53% say that politics is “very important” or“rather important” in their life (69% if “lower class” )

• A role-model effect, however, is unlikely to be immediate– No “effect” on neighboring control states

• The incentive effect for individuals aspiring to run for office in the future islikely a strong explanation

• Assuming that the effects are entirely driven by political aspirations, 28 to 43% ofpotential contestants per seat adjusted fertility due to the limits

• Cannot rule out fear or anticipation of stricter fertility limits in otherdimensions, such as for government jobs

38 / 43

Conclusion

• Local leadership ambitions in India appear to be strong

• Individuals are willing to adjust fertility for a chance to hold political officein the future

• The limits, however, incentivize couples to deviate from their preferredfertility path and shrink the candidate pool

• Unintended sex ratio effects

• The overall effectiveness of the two-child laws thus depends on themagnitude of welfare gains from fertility decline relative to these costs

40 / 43

Conclusion

• Similar fertility limits have been proposed for members of state legislativeassemblies and the national parliament in India

• Rajasthan and Haryana have enacted education requirements for Panchayatcandidates

• Implications for the efficacy of democratic institutions in protecting thewelfare of the socially disadvantaged, who may have higher fertility thanelites due to lower access to contraception and higher risk of child mortality,and depend more on political representation to obtain resources prone toelite capture

• Potentially counteract the beneficial impacts of mandated caste and genderquotas

• Broadly, our results suggest otherwise

41 / 43

Disqualifications during 2000-04

State Number of disqualifications(excluding rejected nominations)

Haryana 1,350Rajasthan 548Madhya Pradesh 1,140Chhattisgarh 766Andhra Pradesh 94*

*Data available for 15 out of 23 districts

Back

42 / 43

Sex ratio of first births

Male = 1 All SC ST OBC Upper(1) (2) (3) (4) (5)

Panel A: Only treatment statesTreatst -0.0081 0.0334 0.0120 -0.0474 0.0081

[0.0110] [0.0332] [0.0445] [0.0401] [0.0217]34,018 5,818 5,783 7,511 14,906

Baseline mean 0.5152 0.5062 0.5103 0.5162 0.5198

Panel B:Treatst -0.0007 0.0325 -0.0063 -0.0304 0.0093

[0.0076] [0.0295] [0.0339] [0.0228] [0.0162]86,023 15,245 9,265 19,345 42,168

Baseline mean 0.5150 0.5128 0.5096 0.5173 0.5158

Panel C:Ts ∗ Postst -0.0042 0.0179 -0.0367 -0.0263 0.0092

[0.0122] [0.0379] [0.0486] [0.0219] [0.0199]Postst 0.0056 0.0226 0.0435 -0.0053 0.0003

[0.0126] [0.0275] [0.0491] [0.0175] [0.0185]N 86,023 15,245 9,265 19,345 42,168Baseline mean 0.5144 0.5224 0.5056 0.5151 0.5128

NOTES: This table reports the coefficients from specifications (3a) and (3b). Standard errors in brackets are clustered by stateand in parentheses are wild-cluster bootstrapped by state. *** 1%, ** 5%, * 10%.

Back43 / 43